American Political Science Review (2019) 113, 4, 883–901

doi:10.1017/S0003055419000352 ©AmericanPolitical ScienceAssociation2019.This is anOpenAccess article, distributedunder the termsof the Creative Commons Attribution licence (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted re-use, distribution, andreproduction in any medium, provided the original work is properly cited.

Who Leads? Who Follows? Measuring Issue Attention and AgendaSetting by Legislators and the Mass Public Using Social Media DataPABLO BARBERA University of Southern California

ANDREU CASAS New York University

JONATHAN NAGLER New York University

PATRICK J. EGAN New York University

RICHARD BONNEAU New York University

JOHN T. JOST New York University

JOSHUA A. TUCKER New York University

Are legislators responsive to the priorities of the public? Research demonstrates a strong corre-spondence between the issues about which the public cares and the issues addressed by politicians,but conclusive evidence about who leads whom in setting the political agenda has yet to be un-

covered.Weanswer this questionwithfine-grained temporal analysesofTwittermessagesby legislators andthepublic during the 113thUSCongress.After employinganunsupervisedmethod that classifies tweets sentby legislators and citizens into topics, we use vector autoregressionmodels to explore whose prioritiesmorestrongly predict the relationship between citizens and politicians. We find that legislators are more likely tofollow, than to lead, discussionofpublic issues, results that hold even after controlling for the agenda-settingeffects of the media. We also find, however, that legislators are more likely to be responsive to theirsupporters than to the general public.

INTRODUCTION

An enduring topic in the study of democraticpolities is how responsive governments are tothe preferences of the public. Twomain lines of

inquiry lead this research: Do politicians respond to theissue priorities of the public (Edwards andWood 1999;Jones and Baumgartner 2004; Neundorf and Adams2018; Sulkin 2005)?And, if so, do they reflect the policypreferences that citizens have on these issues (CaugheyandWarshaw 2018; Page and Shapiro 1983; Soroka andWlezien 2009; Stimson, Mackuen, and Erikson 1995)?Our manuscript focuses on the first of these twoquestions, because while a correspondence betweenpublic andpolitical agendas has definitively been shownto exist, there is still high uncertainty about who leadsand who follows in the agenda-setting process. Evi-dence is even more scant on the important question ofwhich citizens have the strongest ability to set politicalagendas: the general public (Downs 1957), attentivecitizens (Aldrich 1995;Arnold 1990), or politicians’ownsupporters (Egan 2013; Kastellec et al. 2015).

Weaim to shednew light on these classic and relevantpolitical science questions by analyzing the issues towhich members of the US Congress and the Americanpublic pay attention. Although determining whetherpoliticians also follow constituents’ issue preferencesandpriorities on thepolicies they implement—and, if so,which constituents’ issue preferences—is of equal rel-evance to say the least, “policy actions cannot be takenunless attention is directed at the matter” (Jones andBaumgartner 2004, 2). Hence, disentangling whetherpoliticians devote more time discussing an issue afterattention to that issue by the public increases is a first

Pablo Barbera , Assistant Professor, School of InternationalRelations, University of Southern California, pbarbera@usc.edu.

Andreu Casas , Moore-Sloan Research Fellow, Center for DataScience, New York University, andreucasas@nyu.edu.

JonathanNagler , Professor,Wilf FamilyDepartment of Politics,New York University, jonathan.nagler@nyu.edu.

Patrick J.Egan ,AssociateProfessor,Wilf FamilyDepartmentofPolitics, New York University, patrick.egan@nyu.edu.

Richard Bonneau, Professor, Center For Genomics and SystemsBiology, Courant Institute of Mathematical Sciences, ComputerScience Department, and Center for Data Science, New York Uni-versity; and Flatiron Institute, Center for Computational Biology,Simons Foundation, bonneau@nyu.edu.

John T. Jost , Professor, Department of Psychology, New YorkUniversity, john.jost@nyu.edu.

Joshua A. Tucker , Professor, Wilf Family Department of Pol-itics, New York University, joshua.tucker@nyu.edu.

We thankNick Beauchamp, Ken Benoit, DrewDimmery, AndrewEggers, Thorsten Faas, Michael Lewis-Beck, Jennifer Pan, PaulQuirk,MollyRoberts,AnneliseRussell, Gaurav Sood,David Sontag,Dustin Tingley, and John Wilkerson for their helpful comments andsuggestions to previous versions of this paper. We also gratefullyacknowledgefinancial support for theNYUSocialMediaandPoliticalParticipation (SMaPP) lab from the INSPIRE program of the Na-tional Science Foundation (Award #1248077), the Bill and MelindaGates Foundation, the William and Flora Hewlett Foundation, theRita Allen Foundation, the John S. and James L. Knight Foundation,and Intel. In addition, we would like to thank the Gordon and BettyMoore Foundation and the Alfred P. Sloan Foundation for theirsupport of theMoore SloanData ScienceEnvironment, which fundedCasas’ time on the project. Replication files are available at theAmerican Political Science Review Dataverse: https://doi.org/10.7910/DVN/AA96D2.

Received: July 10, 2018; revised: January 31, 2019; accepted: May 13,2019. First published online: July 12, 2019.

883

Dow

nloa

ded

from

htt

ps://

ww

w.c

ambr

idge

.org

/cor

e. IP

add

ress

: 54.

39.1

06.1

73, o

n 25

Aug

202

0 at

07:

56:1

1, s

ubje

ct to

the

Cam

brid

ge C

ore

term

s of

use

, ava

ilabl

e at

htt

ps://

ww

w.c

ambr

idge

.org

/cor

e/te

rms.

htt

ps://

doi.o

rg/1

0.10

17/S

0003

0554

1900

0352

and crucial step toward fully understanding politicalrepresentation in the United States.1

We believe the lack of clear findings on who leadspublic opinion is partly a function of data limitations, astime and issue units available for previous studies didnot allow for sufficiently granular measurement of therelationship between politicians’ and the public’sagendas. Most existing research relies on monthlysurvey data (typically Gallup’s “Most ImportantProblem” [MIP] question) to measure the publicagenda. However, in our 24-hour media environment,politicians and the public are constantly adjusting theissues to which they devote attention, which means thatchanges in attention allocation are likely to happenwithin monthly survey waves. Hence, while survey dataallow us to observe whether the public and politicalagendas covary, they provide limited information onwhichonemovesfirst.Moreover, existinganalyses traceattention to issue categories that are very broad (such as“the economy” or “immigration”), which has the ad-vantage of facilitating comparisons across long periodsof time and units, such as states and countries, but canmake it difficult to uncover who leads public opinion bygrouping together issues that are in fact quite different.2

In an effort to address previous data limitations, wepursue a novel empirical strategy by using themicroblogging social media platform Twitter to mea-sure the expressed agenda of legislators and theAmerican public. To be clear, our goal is not to assesswhether a social media platform such as Twitter isa useful agenda setting tool for politicians or the publicbut rather to use their “tweets” as a proxy to measureattention being paid to political issues. We are reas-sured in pursuing such a measurement strategy be-cause virtually all members of the US Congress areactive Twitter users and their tweets have been shownto constitute a standardized representation of theirexpressed issue agenda (Casas and Morar 2015).Moreover, the issues Americans discuss on socialmedia are highly correlated with other measures ofissue salience such as the MIP survey question(O’Connor et al. 2010).3

Twitter data provide two main advantages to ad-dress the questions at hand. First, the data allow us tomeasure public and political agendas using the samesource: both members of Congress and their

constituents are present on the platform, sendingtweets that have the same format and symbolic ref-erences such as hashtags. Second, the high granularityof the data allows us to observe swiftly changingtemporal patterns in topic salience. We are thereforeable to pinpoint with precision the extent to whichpoliticians allocate attention to different issues beforeor after shifts in issue attention by the public (orwhether they devote attention to the issue at all).Although previous work has used Twitter data toevaluate the issues to which politicians and con-stituents pay attention (particularly, the work by LillyHemphill and colleagues: e.g., Hemphill and Roback(2014) and Shapiro and Hemphill (2017)), this workhas examined a limited number of issues and hasprimarily focusedon issue congruence rather than issueresponsiveness.

We first analyze all tweets sent by members of the113th Congress from January 2013 to December 2014.Using a Latent Dirichlet Allocation (LDA)model, weextract 100 topics that represent the diversity of issueslegislators discuss on the social networking site. Weshow that this method is able to classify legislators’tweets into a set of validated topics that exhibitmeaningful variation over time and across parties. Wethen employ a vector autoregression (VAR) approachto explore the extent to which legislators’ expressedpolitical agendas adapt after a change in issue attentionby three different subgroups of the public: partisans,people who are particularly attentive to politics, anda random sample of US Twitter users. Our tests ex-amine the extent towhich changes in issue attention bythese groups of citizens predict policy makers’agendas.

Our observational analysis is necessarily of a de-scriptive rather than a causal nature. Our VAR ap-proach takes advantage of multiple lags of data todistinguish when groups lead conversations aboutparticular topics and by contrast when they are joiningdebates that are already established, but it cannot ruleout confoundingbyunobservable factors.Nevertheless,our analyses provide revealing information about theconfiguration of political agendas and public opinion inthe United States. Further, it allows us to ascertain inways not previously possible whether observable dataconform to empirical implications of theories specifyinghow the agendas of different sets of actors impact thoseof others, providing corroborative evidence for some ofthese theories.

Our findings show definitively that members ofCongress are more likely to follow the issue prioritiesof the public than to lead them. However, this re-sponsiveness is limited in ways that reinforce polar-ization and inequality. Lawmakers are more likely tochange their behavior after shifts in attention by partysupporters, as previouswork leads us to expect (Bawnet al. 2012; Clinton 2006; Egan 2013; Kastellec et al.2015; Shapiro et al. 1990). To a lesser extent, politi-cians are also responsive to the issue priorities ofattentive citizens over those less inclined to followpolitics (Aldrich 1995;Arnold 1990).But despitewell-established models predicting that politicians should

1 To be clear, we are not, however, addressing the downstreamquestion of whether policy outcomes reflect the preferences of thepublic or particular groups of the public; see Gilens (2012) andGilensand Page (2014) for an analysis of the determinants of policyoutcomes.2 For example, an increase in public attention to the Dakota AccessPipeline followed by Congressional hearings on a fracking bill (e.g.,S.785 of the 114th Congress) would be miscategorized as a case ofagenda responsiveness by a commonly used issue classification inresponsiveness research, the Policy Agendas Project issue classifica-tion (JonesandBaumgartner2004):bothactionswouldbecategorizedinto Natural Gas & Oil (803) within the Energy category. Althoughboth energy related, these two issues are distinct, and assuming thatCongress is reacting and being responsive to a preceding public at-tention change can be misleading.3 We further test this assumption in Online Appendix A.

Pablo Barbera et al.

884

Dow

nloa

ded

from

htt

ps://

ww

w.c

ambr

idge

.org

/cor

e. IP

add

ress

: 54.

39.1

06.1

73, o

n 25

Aug

202

0 at

07:

56:1

1, s

ubje

ct to

the

Cam

brid

ge C

ore

term

s of

use

, ava

ilabl

e at

htt

ps://

ww

w.c

ambr

idge

.org

/cor

e/te

rms.

htt

ps://

doi.o

rg/1

0.10

17/S

0003

0554

1900

0352

reflect the priorities of the general public (Downs1957), we find little evidence for this. Our findings alsosuggest that mainstream media is in part to blame forthis inequality on issue responsiveness: mass mediaare more likely to cover those issues that are of in-terest to partisans, and they often lead the politicalagenda.

POLITICIANS’ RESPONSIVENESS TO THEPUBLIC’S PRIORITIES

Empirical studies on policy (not issue) responsivenesshave substantially advanced due to great innovationsin data collection and measurement (Burstein 2014;Caughey and Warshaw 2018; Gilens 2012; Lax andPhillips 2011; Soroka andWlezien 2009;Tausanovitchand Warshaw 2014), but without a more clear un-derstanding of issue responsiveness, an evaluation ofthe extent to which governments are responsive totheir citizens is incomplete. As Jones and Baum-gartner (2004) note, “How representative is a legis-lative action that matches the policy preferences ofthe public on a low priority issue but ignores highpriority issues?” (p. 2). For politicians to be trulyresponsive to the public, they first need to pay at-tention to the issues constituents deem relevant, andthen their actionsmust reflect people’s preferences onthose issues.

Research on agenda setting and political re-sponsiveness in the United States has found a strongrelationship between the issue priorities of the publicand the agenda of members of Congress (Baumgartnerand Jones 1993). For issues such as the economy, health,environment, and foreign trade, changes in public issuesalience (measured using Gallup’s long-standing MIPquestion) correlate at high levels with changes in po-litical attention (measured as the proportion of Con-gressional hearings on the same issue) (Jones andBaumgartner 2004).

However, existing studies on issue responsiveness donot clearly address a very important question:who leadswhom (Page 1994)? Are policy makers more likely tofollow than to lead changes in issue attention by theirconstituents, or is it the other way around? Researchindicates that both scenarios are possible, but it is un-clear who (if any) has the largest capacity to lead theissue agenda of the other.

On the one hand, research on policy responsivenessargues that politicians have strong incentives to beresponsive to the preferences of the public (Erikson,Mackuen, and Stimson 2002; Geer 1996; Stimson,Mackuen, and Erikson 1995). Building on the “ret-rospective voting” idea, (Campbell, Dettrey, and Yin2010), scholars, such as Stimson, Mackuen, andErikson (1995), argue (and find) that electorally ori-ented politicians update their preferences tomaximizereelectionprospects once theyperceive a shift in publicopinion: “when politicians perceive public opinion tochange, they adapt their behavior to please theirconstituents” (p. 545). Canes-Wrone andShotts (2004)also show that public opinion can influence the

preferences of political figures, such as the president,particularly on issues directly related to people’s dailylife. Overall, this literature suggests that politicians areresponsive to public priorities and leads to the ex-pectation that (H1) the public’s priorities predict theissues towhichmembers ofCongress subsequently payattention.

On the other hand, another body of research arguesthe opposite. Building on the image of “policy-ori-ented” politicians, scholars argue that most politiciansaremainlymotivatedbypolicy goals rather thanby thegoal of seeking reelection (Jacobs and Shapiro 2000).Research shows thatmost citizens are not interested in(Hibbing and Theiss-Morse 2002) and know very little(Delli Carpini and Keeter 1996) about politics, andthat instead of evaluating politicians based on theirpast actions and performance, they make decisionsbased on group attachments (Campbell et al. 1960)and elite cues (Lupia, McCubbins, and Arthur 1998;Sniderman, Brody, and Tetlock 1993). Authors suchas Lawrence R. Jacobs, Robert Y. Shapiro, andBenjamin I. Page draw on this literature and their ownempirical evidence to show that policy-oriented pol-iticians take advantageof people’s political disconnectto set the agenda to their liking. In interviews withadministration officials, the authors are repeatedlytold that the government tracks public opinion “not to‘pander’ but to educate, lead, or otherwise influencepublic attitudes” (Jacobs and Shapiro 1997, 3).Overall, this other body of research leads to the ex-pectation that (H2) members of Congress initiatedebates about issues that are subsequently followedby the public.

Thus, there are good reasons to believe that poli-ticians follow the issue preferences of the public, butalso that the public responds to politicians’ issue pri-orities. But who has the strongest ability to lead theissue agenda of the other? A primary contribution ofour analysis will be to evaluate the magnitude of theseeffects to determinewhohas the largest agenda-settingeffect (if any). We explore this question withouta theoretical preference for either hypothesis, butrather as an open debate that must be addressed totruly evaluate the nature of political responsiveness inthe American democratic system (Burstein 2003; Page1994).

MODELS OF RESPONSIVENESS

Beyond whether politicians or the public have thelargest agenda-setting effect, a second question is alsocrucial for advancing a more complete picture of issueresponsiveness in theUnited States: towhom shouldweexpect members of Congress to be responsive?

Despite a substantial number of studies on the issue,the answer is not as straightforward as one might think.As Burstein (2003, 30) points out, “onemight hope that20 years of research would enhance the credibility ofsome [political responsiveness] theories and reduce thatof others. But this does not seem to have happened.” Inparticular, we observe three main theoretical models to

Who Leads? Who Follows?

885

Dow

nloa

ded

from

htt

ps://

ww

w.c

ambr

idge

.org

/cor

e. IP

add

ress

: 54.

39.1

06.1

73, o

n 25

Aug

202

0 at

07:

56:1

1, s

ubje

ct to

the

Cam

brid

ge C

ore

term

s of

use

, ava

ilabl

e at

htt

ps://

ww

w.c

ambr

idge

.org

/cor

e/te

rms.

htt

ps://

doi.o

rg/1

0.10

17/S

0003

0554

1900

0352

pose three different answers to our question of interest.We call them here theDownsian, theAttentive, and theSupporter models.4

In An Economic Theory of Democracy, Downs(1957) argued that, in a bipartisan democratic system,policy makers interested in reelection should be re-sponsive to the median voter or “centrist opinion”(Jacobs and Shapiro 2000). The implications for po-litical responsiveness are easier to envision froma policy—as opposed to from an issue responsiveness—perspective: members of Congress should adopt thepolicy preference of their median constituent. Fol-lowing the same logic, we should also expect legislatorsto increase their chances of reelection by focusing onissues that a majority of the general public deems rel-evant. Some strongly disagree andargue that politicianshave very little incentives to devote attention to thepreferences of the median voters: only a small pro-portion of the mass public pays attention to, knowabout, and participate in politics; and those who do aremore likely to be partisans than the typical medianvoter (Hibbing and Theiss-Morse 2002; Delli Carpiniand Keeter 1996; de Vreese and Boomgaarden 2006).5

Nevertheless, existing empirical evidence still givessome credit to the Downsian logic. An extensive lit-erature on cue-taking argues that even the least in-formed and attentive public often draws on multiplesources of information to make decisions about poli-tics, also keeping politicians in check (Lau and Red-lawsk 2006; Lupia, McCubbins, and Arthur 1998). Infact, in their exhaustive work on policy responsivenessin the United States and Canada, Soroka and Wlezien(2009) find “roughly the same degree [of policy re-sponsiveness] across groups. In most cases, represen-tation is neither markedly better nor markedly worsewhen we look solely at certain groups” (p. 165).6

Hence, given the conflicting arguments and evidence, itis pertinent to test whether a Downsian logic is still inplace.Amain testable hypothesis that derives from theargument is that (H3) changes inattention allocationbythe general public predict changes in issue attention bymembers of Congress.

Other scholars, however, disagreewith this premiseas being too optimistic about the public’s agenda-setting role. Instead of responding to the medianvoter or the general public, some believe members ofCongress have incentives to be mostly responsive toattentive voters. Studies of opinion formation showthat most voters do not follow day-to-day politics(Hibbing and Theiss-Morse 2002) and that many donot have clear issue priorities nor policy preferences

(Converse 2006). Nevertheless, this is not the case forall citizens. Some attentive voters care a great dealabout the political world, and according to theoreticalmodels such as Katz and Lazarsfeld (1955)’s “two-step communication flow” and Page and Shapiro(1992)’s “rational public,” these attentive voters havethe potential to influence the issue priorities andpreferences of less attentive citizens. This type of logicleads congressional scholars such as Arnold (1990)and Aldrich (1995) to argue that members of Con-gress should be particularly concerned about theissues to which attentive voters pay attention. Atestable hypothesis that derives from this logic is that(H4) changes in attention allocation by attentivepublics predict allocation changes by members ofCongress.

Another group of researchers proposes a third al-ternative: legislators should be mostly interested inresponding to core party supporters. They have issuepriorities that are easier to distinguish and represent(Wright 1989), they play a very active role in nomi-nation processes (Bawn et al. 2012), their support iscrucial to win not only primaries (Fenno 1978; Gerberand Morton 1998) but also general elections (Hol-brook andMcClurg 2005), and the priorities of policy-oriented members are more likely to align with theirs(Egan 2013; Kastellec et al. 2015). Some empiricalresearch finds that in fact legislators are more likely torepresent the policy preferences of their supporters(Clinton 2006; Kastellec et al. 2015; Neundorf andAdams 2018; Shapiro et al. 1990), but no research yetexists showing whether that is the case for issue at-tention allocation. From this model, however, we canderive that (H5) changes in attention allocation byparty supporters predict allocation changes by mem-bers of Congress.

As a final theoretical consideration, scholars argue(andfind) that themore salient an issue becomes in theeyes of the public, the larger the degree of politicalresponsiveness we should expect (Burstein 2003;Canes-Wrone and Shotts 2004; Jones 2004; Sorokaand Wlezien 2009; Sulkin 2005): “the public impor-tance of policy domains may tell us a lot about policymakers’ responsiveness. There is good reason, afterall, to expect policymakers to reflect the importanceofthe different domains because of possible electoralconsequences” (Wlezien 2004, 7). If politicians aim tobe responsive to a certain group of the public, theyshould be interested in reacting to shifts in attentioninvolving issues that are particularly salient to thatgroup. In other words, if members of Congress aremainly responsive to their party supporters (or to theattentive or general public), they should be morelikely to react to shifts in attention by party supporterson issues that take on 10% rather than 1% of theirsupporters’ discussion. Hence, building on this liter-ature, if the previous responsivenessmodels apply, weshould expect that (H6) to the extent that particularissues are more salient among the general public(according to the Downsian model), attentive citizens(according to the Attentive model), and party sup-porters (according to the Supporter model), these

4 There is a fourthmain theoretical model that for data limitations weare unable to test in this paper: the argument that policy makers areresponsive to wealthier constituents (Gilens 2012).5 DelliCarpiniandKeeter (1996), forexample, found thatonly30%ofAmericans know the name of at least one of the two Senators rep-resenting their State.6 Theauthorsdofind, however, thatwhendiscrepancies exist betweenthe policy preferences of some groups, politicians do tend to be moreresponsive to somegroups: thosewith ahigher income, education, andpolitical sophistication.

Pablo Barbera et al.

886

Dow

nloa

ded

from

htt

ps://

ww

w.c

ambr

idge

.org

/cor

e. IP

add

ress

: 54.

39.1

06.1

73, o

n 25

Aug

202

0 at

07:

56:1

1, s

ubje

ct to

the

Cam

brid

ge C

ore

term

s of

use

, ava

ilabl

e at

htt

ps://

ww

w.c

ambr

idge

.org

/cor

e/te

rms.

htt

ps://

doi.o

rg/1

0.10

17/S

0003

0554

1900

0352

publics’ priorities will more strongly predict the pol-iticians’ agenda.7

THE MECHANISMS OF ISSUERESPONSIVENESS

Here, we use messages sent on Twitter by politiciansand the public as measures of these individuals’ issuepriorities, allowing us to evaluate the extent to whichreciprocal relationships exist among these priorities.Previous research has suggested severalmechanisms bywhich politicians might become informed about thepublic’s priorities, and vice versa. In particular, thisliterature puts forward a set of mechanisms throughwhich the issue priorities of politicians and the publiccan directly lead the priorities of the other, as well asways in which the mass media can channel these re-ciprocal dynamics.

One mechanism by which politicians and citizenslearn about one another’s priorities is direct interactionand communication between lawmakers and theirconstituents, which include town halls, “lobby days”organized by interest groups and offline and onlinecorrespondence.Forexample, about24%ofAmericansreport having written a letter to a public representative(Schlozman, Verba, and Brady 2012). Surveys indicatethat politicians pay some attention to social mediamessages from constituents as well, although thesemessages are not weighed as heavily as other constit-uent communications (Chen, Lee, and Marble 2018).Another way that politicians learn about citizens’ pri-orities is by tracking public opinion. Polling organ-izations, such as Gallup and the Pew Research Center,regularly release polls revealing the public’s issue pri-orities. Political parties, campaign staff, and govern-ment agencies run their own polls to assess the issuesthat are of interest to the public (Jacobs and Shapiro2000). Modern practices also include using dashboardsto track the issues that aremostlydiscussedby citizens insocial media (Webster and Ksiazek 2012).

In this study,we focusonwhat is arguably theprimarymechanism by which politicians and the public learnabout each other’s issue priorities: the news media.First, the media’s powerful role in this regard derivesfrom the fact that lawmakers are incentivized to expendeffort to generate media coverage of their priorities,while at the same timemedia outlets are incentivized tocover issues that resonate with their audiences’ prior-ities. A wide range of factors determine media content,including assigned “beats,” journalistic practices, andthe occurrence of newsworthy events (Graber 1997;Shoemaker and Reese 1996). Among these factors,audience preferences and political institutions playa very important role, ensuring that media covers issuesthat are of interest to both the public and politicians.And second, the media’s substantive role derives from

its ability to drive both public and political attention(McCombs and Shaw 1972; Zaller 1992).

The media is responsive to political elites in partbecause political institutions represent an importantsource of constant newsworthy information (Shoe-maker and Reese 1996). Media outlets regularly ap-point correspondents to institutions such as the WhiteHouse and Congress, ensuring that major issues dis-cussed in thesepolitical venuesachievemedia attention.Media outlets are responsive to public demands in partbecause ofmarket pressures: particularly in a context inwhich most outlets face economic hardships, discussingthe issues that areof interest to thepublic increases theirchances of getting the readers and viewers needed togenerate profits. More than ever, media outlets todayhave a wide range of instruments in their hands tomeasure, and respond to, the issues inwhich thepublic isinterested (including tracking social media attention)(AnandandPeterson 2000;Webster andKsiazek2012).In addition, public sentiment is often reflected bynewsworthy political events in themselves, includingelection results, strikes, and demonstrations (Gitlin1980).Tobe sure, neither thepublic nor politicians haveexclusive control over the agenda, which is frequentlyset by external unexpected shocks (such as naturaldisasters) (Birkland 1998) and recurrent events (andexpiring statutory provisions) (Adler and Wilkerson2012) that simultaneously affect the attention distri-butionofpoliticiansbut alsoof themedia and thepublic.

There is clear evidence showing that not only poli-ticians and the public lead media attention but also thatthemedia candrivepolitical agendas andpublic opinion(Soroka 2002; Walgrave, Stuart, and Nuytemans 2008;Walgrave and Van Aelst 2006). The media “construct”and highlight problems for politicians to solve, and theyincrease the salience of issues that votersmight considerrelevant, which reelection-seeking politicians shouldaddress in other to please them (Wouters andWalgrave2017). Moreover, a long-standing literature also showsthat the issues covered in the media are very likely tolead public opinion and preferences (Boydstun 2013;Iyengar and Kinder 2010; McCombs and Shaw 1972;Zaller 1992).

Exploring all the mechanisms through which groupsof the public and politicians influence the agenda of theother is outside the scope of this study. Nevertheless,given that research points to mass media as playinga crucial agenda setting role, in our analysis we willcontrol for potential media effects as well as explore theextent to which mass media coverage favors particularresponsiveness models.

DATA

Members of Congress on Twitter

To test our hypotheses, we use tweets sent by membersof the 113th House and Senate of the US Congress(2013–14). Twitter use in Congress has increasedsteadily over the past years (Chi andYang 2011; Evans,Cordova, and Sipole 2014; Golbeck, Grimes, and

7 Aswillbeexplainedbelow in the IssueAttentionCongruence section,wemeasure the salience of an issue by calculating the average relativedaily attention that different groups of the public paid to each politicalissue.

Who Leads? Who Follows?

887

Dow

nloa

ded

from

htt

ps://

ww

w.c

ambr

idge

.org

/cor

e. IP

add

ress

: 54.

39.1

06.1

73, o

n 25

Aug

202

0 at

07:

56:1

1, s

ubje

ct to

the

Cam

brid

ge C

ore

term

s of

use

, ava

ilabl

e at

htt

ps://

ww

w.c

ambr

idge

.org

/cor

e/te

rms.

htt

ps://

doi.o

rg/1

0.10

17/S

0003

0554

1900

0352

Rogers 2010; Shapiro, Hemphill, and Otterbacher2012). Ninety-five percent of legislators that served inthe113thCongress hadactiveTwitter accounts, sendinga total of 651,116 messages (excluding retweets), about900 tweets per day.8

Golbeck, Grimes, and Rogers (2010) argue thatmembers ofCongress useTwitter primarily to advertisetheir policy positions and to provide information abouttheir activities. However, more recent studies haveshown that the platform can also be a tool for membersof Congress to be responsive to their constituents, ex-ercise control of the political agenda, express partyloyalty, engage in partisan taunting, and report on theirconstituency service (Hemphill, Otterbacher, andShapiro 2013; Evans, Cordova, and Sipole 2014;Russell2018b, 2018a). Moreover, research indicates that thetopics discussed in tweets are a fair representationof thelegislators’ overall expressed agenda: there is a veryhigh correlation between the issues they discuss onsocial media and their press releases, for example(Casas and Morar 2015).

Citizens on Twitter

In addition to tweets sent by members of Congress, wealso collected tweets sent by different samples ofTwitter users. These allow us to test our hypotheses(H3,4,5,6) regarding the part of the public whose shifts inattention politicians are more likely to follow. Weconsider four samples of Twitter users:

1 General Public: includes about 25,000 Twitter users,sampled by generating random numeric user IDs, thenchecking whether the users existed, and then checkingwhether the users resided in the United States.9

2 Attentive Public: a randomly generated sample ofTwitter users that follow at least one of fivemajormediaoutlets in the United States (CNN, Wall Street Journal,

New York Times, Fox News, and MSNBC). We applya geographic restriction basedon the time zoneonusers’profiles, which is available for most users. In particular,we exclude users whose time zone indicates they arelikely to be located outside the United States. We alsofilter basedonactivity: only userswhohaveever sent 100tweets ormore are included.After applying these filters,the final sample size is 10,000 users.

3 Republican Supporters: a random sample of 10,000Twitter users who follow three or more Republicanmembers of Congress and no Democrat in Congress.The same geographic and activity filters as in the at-tentive public sample are applied here. In Online Ap-pendix F, we demonstrate why this sampling method isable to select party supporters.

4 Democratic Supporters: a random sample of 10,000Twitter users who follow three or more Democraticmembers of Congress and no Republican in Congress.The same geographic and activity filters as in the sampleof Republican supporters apply.

After identifying these four samples, we then col-lected all the tweets they sent during our period ofanalysis (January 2013 to December 2014) usingTwitter’s REST API.10 The final number of users andtweets in each group is available in Table 1. Retweetsare excluded from our sample to avoid inflating thecorrelations we observe between politicians and thepublic regarding the issues they discuss.

Media

As discussed previously, it may well be that both publicand political issue agendas are led by the mass media(Gerber, Dean, and Bergan 2009; Habel 2012; King,Schneer, and White 2017; Ladd and Lenz 2009), par-ticularly soonsocialmedia(Feezell 2018).Toaccount for

TABLE 1. Description of the Tweets in the Dataset

Group N Avg Min Max Tweets

House Republicans 238 1,215 70 8,857 267,311House Democrats 207 1,177 113 5,993 222,491Senate Republicans 46 1,532 73 6,627 67,412Senate Democrats 56 1,616 150 10,736 87,307

Random sample 25k 465 1 8,926 11,316,396Informed public 10k 948 100 5,861 9,487,382Republican supporters 10k 1,091 100 8,804 10,911,813Democratic supporters 10k 1,306 100 5,122 13,058,947

Media outlets 36 7,803 8 15,858 273,121

Note: Periodof analysis: January 1, 2013, toDecember31, 2014.N corresponds to thenumberof Twitter accounts in eachsample.Avg,Min,andMax correspond to the average, minimum, andmaximum number of tweets, respectively, sent by individual users in each group duringthe whole period of analysis. Tweets corresponds to the total number of tweets sent by all users in each group during the period of analysis.

8 SeeOnlineAppendixF for informationregarding thedatacollectionprocess.9 A description of the procedure we followed to build this sample canbe found in Online Appendix B.

10 We conducted the data collection for the random sample (GeneralPublic group) in 2015 and were constrained by Twitter’s rate limits,whichdonotallowdownloadingall ofuser’s tweets. In this case, partofthe tweets sent in 2013 by about 44% of the random users are notincluded in the sample.

Pablo Barbera et al.

888

Dow

nloa

ded

from

htt

ps://

ww

w.c

ambr

idge

.org

/cor

e. IP

add

ress

: 54.

39.1

06.1

73, o

n 25

Aug

202

0 at

07:

56:1

1, s

ubje

ct to

the

Cam

brid

ge C

ore

term

s of

use

, ava

ilabl

e at

htt

ps://

ww

w.c

ambr

idge

.org

/cor

e/te

rms.

htt

ps://

doi.o

rg/1

0.10

17/S

0003

0554

1900

0352

this possibility, we also collected tweets from a sample ofmedia outlets and use them to control for media effects.In particular, we collected all tweets sent over the sametime period from the Twitter accounts of the 36 largestmedia outlets in the United States (print, broadcast,online), as identified by the Pew Research Center.

MEASURING ATTENTION TO POLITICALISSUES WITH TOPIC MODELS

Our purpose in this paper is to characterize the differentissues that members of Congress, ordinary citizens, andmedia outlets discussed on Twitter, and how their im-portance varies over time and across groups defined bytheir partisanship andpolitical interest. To extract thesecategories, we estimate a probabilistic model of wordoccurrences in documents called anLDA(Blei, Ng, andJordan 2003), which belongs to a general category oflatent variablemodels that infer topics from documentsusing a “bag-of-words” approach.

As we explain in greater detail in Online AppendixG.1, this method treats each document as a randommixture over latent topics, and each topic as a proba-bility distribution over tokens. In our analysis, tokensare n-grams (combinations of one and two words). Ourdefinition of “document” is the aggregated total oftweets sent by members of Congress each day, by partyand chamber.11

The alternative to using an unsupervised topic modelwould be for the analyst to choose the topics and thenbuild a supervised classifier predicting them.Despite theexistence of well-known categories of political issues,12

training an accurate classifier would be an incrediblyarduous task, given the large number of categories,making unsupervised models a preferable option.However, as we show in Online Appendix E.2, it ispossible to map the topics derived from the data to anexisting classification of political topics—the topics usedinJonesandBaumgartner (2004)—withsimilar results.13

And as we demonstrate in Online Appendix G.3, thetopics generatedby themodel pass standard tests of bothpredictive and semantic validity (Quinn et al. 2010).

Note two additional features of our analysis. First, wefit the model at first only for members of Congress (in-stead of fitting it to the messages sent by all groups) toincrease the likelihood of discovering topics that werepolitically salientduring the113thCongress,andthenusethe estimated parameters to compute the posterior topic

distributions for citizens and media outlets, also aggre-gated by day, based on their observed words. However,a replication of the results based on an LDAmodel fit tothe tweets of politicians, the media, and the public leadsto similar conclusions (see Online Appendix E.4). Sec-ond, in our estimationwe assume that topic distributionsare independent over time, and that thenumberof topicsand the content of each topic is constant over time.

Wefix thenumber of topics toK5100 after exploringa wide range of values by running 10-fold cross-validations and computing common goodness-of-fitmeasures (Chang et al. 2009) (see Online AppendixG.2 for a detailed description of how we chose K).

In general, we find that most of the 100 resultingtopics canbeeasily labeled.However, not all of themarepolitical in nature: for example, we find topics aboutanniversaries and celebrations (Valentine’s Day, FlagDay,ConstitutionDay,Thanksgiving, etc.).Becauseweare not interested in these topics, in our analysis we willonly include political issues, of which 53 were identi-fied.14 After reviewing their content, we noted thatsome topics that referred to a single issuewere classifiedas different topics because distinct words were beingused by different groups when talking about the sameissue. For example, we found separate topics for Re-publican and Democratic members of Congress dis-cussing the 2013 Government Shutdown. This mayinfluence our results by overestimating how oftenparties in Congress respond to their supporters. Toavoid this potential source of bias, we decided to mergesome topics and focus our analysis on 46 political issues.Table 2 displays the list of all these topics we haveclassified as political issues.

RESULTS

Issue Attention Congruence

The key substantive question we want to answer iswhether thedistributionof topics discussedbymembersof Congress leads or follows that of their constituents,and vice versa. Are members following their con-stituents? And if so, are they following particular typesof constituents?

Similar to previous studies on the issue, we start byexamining simple congruence in the way members ofCongressandcitizensallocateattention to the46politicalissueswe identified. In this issue congruence framework,a correlation between the public and the political agendais a necessary condition for political responsiveness to bepresent. Table 3 displays Pearson correlation coef-ficients, indicating how similar the issue distribution of

11 There are two reasons for this decision. First, LDA assumes thateach document is a mixture of topics, which is appropriate for ourconceptualizationof eachday’s tweets as thepolitical agenda that eachparty within each legislative chamber is trying to push for thatspecific day. Second, conducting an analysis at the tweet level iscomplex, given its very limited length. The existing literature on topicmodeling of tweets has found that applications that aggregate tweetsby author or day outperform those that rely on individual tweets(Hong and Davison 2010).12 Such as the ones from the Comparative Agendas Project orManifesto Project.13 Forageneraloverviewof theuseof text-as-datamethods inpoliticalscience research, seeGrimmer and Stewart (2013) andWilkerson andCasas (2017).

14 To identify the list of relevant political topics, five coders used theinformation contained in our topic dashboard (pablobarbera.com/congress-lda) to classify each of them into three categories: non-political topics (e.g., Valentine’s Day), political topics but not relatedto issues (e.g., public communication by House Republicans), andpolitical issues (e.g., gun violence, government shutdown, obama-care). Average intercoder agreement was 83% and Cronbach’s alphawas 0.92. We chose as political issues those where the modal classi-fication (three or more coders) agreed to classify as such.

Who Leads? Who Follows?

889

Dow

nloa

ded

from

htt

ps://

ww

w.c

ambr

idge

.org

/cor

e. IP

add

ress

: 54.

39.1

06.1

73, o

n 25

Aug

202

0 at

07:

56:1

1, s

ubje

ct to

the

Cam

brid

ge C

ore

term

s of

use

, ava

ilabl

e at

htt

ps://

ww

w.c

ambr

idge

.org

/cor

e/te

rms.

htt

ps://

doi.o

rg/1

0.10

17/S

0003

0554

1900

0352

Democratic and Republican supporters, attentive pub-lics, and thegeneral public are to theexpressedagendaofRepublicans and Democrats in Congress over the two-year period studied. Higher coefficients indicate thatgroups tend to discuss the same issues.

These initial results show potential for corroboratingthe presence of political responsiveness at the issueattention level, and they seem to indicate that someresponsiveness models have a stronger explanatorypower than others. In particular, these results providestronger support for the Supporter and, to a lesserextent, the Attentive models, than for the Downsianargument. There is a positive, and in some cases large,correlation between the agenda of members of Con-gress and the issues discussed by their constituents.Nevertheless, when paying attention to the coefficientsfor specific groups, we observe the highest correlationsto bebetweenmembers and their party supporters (0.69for Democrats and 0.77 for Republicans) and betweenmembers and the attentive public (0.49 for Democrats

and 0.52 for Republicans). The correlation between theexpressed agenda of legislators and the attention allo-cation of supporters of the other party is much lower (a0.41 correlation between Democratic members andRepublican supporters, and 0.51 between Republicanmembers and Democratic supporters). We observe thelowest correlation coefficients when comparing theagenda of lawmakers and the issues the general publicdiscuss. As expected, the issue attention distribution ofthe media is also highly correlated to the agenda ofmembers of Congress, emphasizing mass media’sagenda-setting and mediating role.

In Figure 1, we provide information about the av-erage daily attention that each party in Congress, eachpublic group, and the media paid to the political issuesunder study. This figure provides a more detailed un-derstanding of the agenda level correlationswe observein Table 3 and some potential reasons as to why weobserve a particularly strong relationship between theissue agenda of members of Congress and the attention

TABLE 2. List of Political Issues

Topic Number Label Topic number Label

3 Investigation of Benghazi attack 50 Climate change7 100 days of #BringBackOurGirls campaign 51 Lame duck congress9 Gender wage gap 53 Minimum wage12 Republican issues Spring 2013 58 Affordable Care Act14 Marriage equality 62 Border crisis in Texas15 Gun violence 63 Obamacare (employer mandate)16 Abortion (pro-life) 64 FAA furloughs cause flight delays18 Veteran affairs delays scandal 66 Malaysia Airlines crash in Ukraine20 NSA surveillance scandal 67 Comprehensive immigration reform23 #BringBackOurGirls campaign 70 #MiddleClassFirst campaign28 Employment Non-Discrimination Act 75 Military Justice Improvement Act32 Islamic state 81 Poverty (SNAP program)33 Use of military force in Syria 83 Twenty-first century cures initiative36 Ebola 85 Unemployment insurance37 Social security 88 IRS scandal39 Keystone XL pipeline 89 Obamacare (website and implementation)41 Immigration (border security) 93 Jobs bills omnibus43 Executive action on immigration 96 Violence Against Women Act46 Unemployment numbers reports 97 Protests in Ukraine and Venezuela47 Paul Ryan budget proposal 99 CIA detentions and interrogations report48 Black history month 100 #ObamacareInThreeWords campaign(101) Student debt (102) Hobby lobby supreme court decision(103) Budget discussion (104) 2013 government shutdown

Note: The topic number in parentheses indicate issues that have been created ad hoc by merging very similar topics from the topic model.

TABLE 3. Correlation in Issue Attention BetweenMembers of Congress andGroups of the Public andthe Media Over 46 Political Issues

Group Democrats in Congress Republicans in Congress

Democratic supporters 0.69 0.51Republican supporters 0.41 0.77Attentive public 0.49 0.52General public 0.38 0.34Media 0.52 0.63

Pablo Barbera et al.

890

Dow

nloa

ded

from

htt

ps://

ww

w.c

ambr

idge

.org

/cor

e. IP

add

ress

: 54.

39.1

06.1

73, o

n 25

Aug

202

0 at

07:

56:1

1, s

ubje

ct to

the

Cam

brid

ge C

ore

term

s of

use

, ava

ilabl

e at

htt

ps://

ww

w.c

ambr

idge

.org

/cor

e/te

rms.

htt

ps://

doi.o

rg/1

0.10

17/S

0003

0554

1900

0352

distribution of their party supporters. We see, for ex-ample, how Democrats in Congress and Democraticsupporters paid much more general attention to theAffordable Care Act (row 3 of figure 1) and MarriageEquality (row 15) than did Republicans, whereasRepublicans in Congress and Republican supporterspaid more attention than did Democrats to the (trou-bled) release and implementation of the ACA website(11th to the last row), its employer mandate clause (lastrow), and to the Border Crisis in Texas (10th to the lastrow). The attentive public, and especially the generalpublic, paid less attention not only to these issues butalso to all political issues in general. (See Online Ap-pendix G.4 for further discussion of these results.)

Who Leads? Who Follows?

Theprevious correlations andpercentages, however, arenot sufficient evidence to conclude that members ofCongress lead the issue attention distribution of their

constituents nor to adjudicate between the competingDownsian, Attentive, and Supporter models. Here, wetake advantage of the time series nature of our dataset toestablishwhoputs issueson theagendafirstbyestimatingaVARmodelswith topic-fixedeffects.Thesemodels arewell-suited to capture the relationship between endog-enous variables (Freeman,Williams, and Lin 1989; Sims1980) and have been used in previous political sciencestudieswith similar objectives (Edwards andWood1999;Enders and Sandler 1993; Wood and Peake 1998).15

FIGURE 1. Average Issue Attention by Groups of Politicians, the Public, and the Media

Note: Attention is represented as daily posterior LDA topic probabilities expressed in percentages. These are percentages based on all 100topics of the LDA model.

15 In this paper, we do not control for the agenda of the President fortwo main reasons. First, Edwards and Wood (1999) do not find thePresident to have a relevant agenda-setting capacity, and second, themethod we employ in this paper to measure agendas (proportion ofdaily tweets on a set of issues) is well suited to measure group but notindividual agendas, given that single individuals do not tweet fre-quently enough tobuildunbiasedmeasures.Needless to say, revisitingthe question of agenda-setting power of the President in the era ofTrump would likely be a good idea for future research.

Who Leads? Who Follows?

891

Dow

nloa

ded

from

htt

ps://

ww

w.c

ambr

idge

.org

/cor

e. IP

add

ress

: 54.

39.1

06.1

73, o

n 25

Aug

202

0 at

07:

56:1

1, s

ubje

ct to

the

Cam

brid

ge C

ore

term

s of

use

, ava

ilabl

e at

htt

ps://

ww

w.c

ambr

idge

.org

/cor

e/te

rms.

htt

ps://

doi.o

rg/1

0.10

17/S

0003

0554

1900

0352

In our VAR model, we have a set of stationary timeseries Yi representing the proportion of daily attentioneach of our groups i16 paid to each topic j in day t of the113th Congress.17 The values of these random variablesrange from 0 to 1 but neither of the extreme values arepresent (0,Yijt, 1).Theirdistributionsareright skewed,with fewdays of very high issue attention andmuch lowerattentionduring the restof the two-yearperiod.Wefollowa common practice in time series analysis of skewedproportions (Wallis 1987)andmodel the logoddsZiof thedescribed series Yi instead of the raw proportions.

We then express the autoregressive and endogenousrelationship of these variables as a system of equationsinwhich each variableZi is a function of its previous lagsplus the lags of the other variables. Given that there areno time restrictions when it comes to posting messageson Twitter, we would theoretically expect members ofCongress to follow changes in public issue attentionquite rapidly. However, to account for the potential oflonger-term decay we use a seven-lag structure.18 Thefinal model can be & formally expressed as follows:

Z ¼ logY

1� Y

� �

Zi;j;t ¼ aj þ�i�7

p¼1bi;pZi;j;t�p þ «i;j;t:

(1)

Note that given the issue-fixed effects structure of themodel (aj), we are assuming that the estimates of in-terest are constant across issues. Although this is aninaccurate assumption, it is a useful one for what weintend to accomplish here. It allows us to estimate howmuch on average we should expect changes in issueattention by a given group to predict subsequent at-tention allocation of the other groups.19

The results of the estimated VARmodel can be bestexpressed using cumulative impulse response functions(IRFs). These cumulative IRFs indicate how an x-unitincrease in attention to a given topic by a group predictsthe cumulative attention that other actors dedicate tothe same topic over time. Cumulative IRFs can becalculated for a varying number of subsequent days.Wecalculate and report in Figure 2 two different types ofIRFs fora15-dayperiod.20 Inbothcases,weassumethat

at day 0 none of the groups is paying attention to a givenissue j. First, we want to explore the effect of briefchanges in attention and we calculate how a 10 per-centage point increase in attention to an issue by eachgroup (going from 0% to 10% of attention in day 0)affects future issue attention by the other groups. Weare also interested in the effect of attention changes thatlast longer and calculate how a permanent attentionchange to a given issue from 0% to 10% by one groupaffects the attention of the others.21Each single panel inFigure 2 shows howmuchmore cumulative attention tothe issue the group in the panel title is predicted to payafter a one-time (in gray) and permanent (in black) 10percentagepoint increase by the groups along the y-axis(the row groups) 15 days ago. The predicted responses(95% confidence interval lines) are expressed in per-centagepoints (0–100 scale).Most one-time effects (graycoefficients) range from 0 to 4. We believe these aremeaningful and substantive responses.Asweknow fromthe long-standing literature on agenda setting, politics isoften a fight for attention and simply getting an issue intotheagenda isextremelydifficult (JonesandBaumgartner2005; Schattschneider 1975). Moreover, dynamics ofattention often follow nonlinear functions with tippingpoints, where small amounts of additional attentionaround the tipping point can have large political impli-cations (Baumgartner and Jones 1993; Kingdon 2013).

The results in Figure 2 corroborate the first twoexpectations in regard to the ability of members ofCongress and the public to predict each other’s issueattention. Politicians fromboth parties (first two rows inall the panels inFigure 2) are able topredict the attentiondistribution of the public (H2). Specifically, they are ableto lead the issue attention of party supporters and at-tentive publics, although both parties lead the issue at-tention of their own supporters than supporters of theopposing party.Andwe see in the far right panel that theissues prioritized by both parties appear to be very poorpredictors of the issue attention of the general public.

We also find strong evidence supporting a politicalresponsiveness dynamic (H1): we see changes in issueattention by citizens to be positive predictors of theissues members of Congress discuss, and we also seethese effects to always be of a largermagnitude than theability ofmembers ofCongress to lead the agendaof thepublic. The ability of Republican supporters to setpolitical issue agendas represents the most extremecase.This groupof thepublic is predicted to increase thecumulative amount of attention to an issue only by 0.75and 1.25 percentage points 15 days after a 10-pointincrease in attention by Democrats and Republicansin Congress, respectively (first and second gray esti-mates from the top in the fourth panel from the right).

16 Democratic and Republican members of Congress, Democratic andRepublican Supporters, Attentive Public, General Public, and Media.17 We run Augmented Dickey–Fuller unit root tests and confirm thatthe series are stationary.18 Autocorrelation and partial autocorrelation functions varydepending on the group and issue series one explores. However, onaverage we observe autocorrelations to go below 0.1 after 5–9 days,andpartial autocorrelations tobebelowthis level after3–5days;whichindicates that by using a seven-lag structure we are accounting andcontrolling for the autocorrelation nature of these variables.19 As we point out in the Discussion, we believe that future workshould focus on studying how the effects presented here are condi-tional on the issue or issue type at hand. The methods advanced herecan, for example, beused to clearly test hypotheses on the relationshipbetween issue responsiveness and issue ownership. This is not,however, the immediate purpose of this study.20 In Online Appendix D, we provide results based on shorter andlonger term IRFs.

21 When calculating the IRFs for the permanent 10-percentage-pointchange,we insert anew increaseof attentioneveryday to thecovariateof interest until the predicted attention for that group reaches 10%without the needof any extra shock. If we letf represent the attentionincrease we introduce in the covariate, and y the resulting predictedvalueattention for that samecovariate, thenwecan formally expressf

during the 15 day period as follows: fi;t ¼10; if t is 010� yi;t if t is > 0

�.

Pablo Barbera et al.

892

Dow

nloa

ded

from

htt

ps://

ww

w.c

ambr

idge

.org

/cor

e. IP

add

ress

: 54.

39.1

06.1

73, o

n 25

Aug

202

0 at

07:

56:1

1, s

ubje

ct to

the

Cam

brid

ge C

ore

term

s of

use

, ava

ilabl

e at

htt

ps://

ww

w.c

ambr

idge

.org

/cor

e/te

rms.

htt

ps://

doi.o

rg/1

0.10

17/S

0003

0554

1900

0352

The changes in the opposite direction, however, are twoto three times larger: both Democrats and Republicansin Congress are predicted to increase their cumulativeattention by about 3 percentage points (Republicansupporters’ gray estimates in the two most left panels).

These differential effects can be better appreciated inFigure 3, where we rearrange the one-time attentionchanges estimates (gray responses) from Figure 2 tomore easily compare who has the largest ability to leadthe issue agenda of the other, members of Congress orthe public. However, there is an additional factor weneed to take into consideration when interpretingthese results. As we observed in Figure 5 and TableA.6, politicians devote more attention to political

issues than their party supporters (andmuchmore thanthe attentive and general public). This means thatalthough an increase in attention of the same size hasa larger effect when it goes from the public to politi-cians, we are more likely to observe members ofCongress, rather than the public, tomake large shifts inpolitical attention.

Our results also provide strong evidence in favor oftheSupportermodel of responsiveness (H5). Ifwe focusonly on the variables predicting the agenda of membersof Congress the most (two left panels in Figure 2), weobserve that the strongest predictors of a positive at-tention change by lawmakers is a change of attention bytheir own party supporters. The VAR model predicts

FIGURE 2. 15 Day Cumulative IRFs: Predicted Issue Responsiveness Across Groups

Note: The coefficients (with 95%confidence intervals) indicate (in percentage points) howmuchmore cumulative attention the groups in thepanel titles paid to a given issueas a result of the groups in the y-axis increasing the attention to the same issueby10percentagepoints once(in gray) and permanently (in black) 15 days ago.

FIGURE 3. Politicians’ Ability to Set Public Agendas versus the Ability of the Public to InfluencePolitical Agendas

Note: The coefficients (with 95%confidence intervals) indicate (in percentage points) howmuchmore cumulative attention the groups in thepanel titles paid to a given issue as a result of the groups in the y-axis increasing the attention to the same issue by 10 percentage points (inblack) 15 days ago. The gray coefficients indicate the vice versa effect.

Who Leads? Who Follows?

893

Dow

nloa

ded

from

htt

ps://

ww

w.c

ambr

idge

.org

/cor

e. IP

add

ress

: 54.

39.1

06.1

73, o

n 25

Aug

202

0 at

07:

56:1

1, s

ubje

ct to

the

Cam

brid

ge C

ore

term

s of

use

, ava

ilabl

e at

htt

ps://

ww

w.c

ambr

idge

.org

/cor

e/te

rms.

htt

ps://

doi.o

rg/1

0.10

17/S

0003

0554

1900

0352

Democrats in Congress to go from paying no attentionto an issue to dedicating a cumulative attention of ap-proximately 3% as a result of a one-time 10 point at-tention shift by Democratic supporters, and 7% asa result of permanent 10 point change by their partysupporters (Democratic supporter estimates in the leftpanel). We see Republicans in Congress respond sim-ilarly to changes in attention by their own supporters(Republican supporter estimates in the second panelfrom the left). All the other IRFs for the one-time andpermanent attention shocks are of smaller magnitude.

We also find some support for the Attentive model(H4). For example, after a one-time and a permanent 10percentage point change in attention by the attentivepublic, Democratic members of Congress are predictedto increase their cumulative attention by 2.25 and 5.75percentage points, respectively, and Republican policymakers byabout1.75and4.25. Ifwe treat the supportersof theother party also as anattentive public (they follownot one but at least three members of Congress inTwitter), we observe a similar pattern. Changes in at-tention by Democratic Supporters are also predicted tohave a positive effect of 2 and 4.25 points onRepublicanmembers, and changes by Republican Supporters arepredicted to increase the cumulative attention ofDemocratic lawmakers by 2.5 and 5.5 points. However,the estimated effects are of smaller size than the effectswe observed in favor of the Supporter model.

Finally, the results show weak support for theDownsian model (H3). Democratic members of Con-gress are only predicted to increase their cumulativeattention to an issue by 2 and 4.5 percentage points aftera one-time and a permanent 10 point increase of at-tention by the general public. TheRepublicanmembers’response is expected to be even lower—their cumulativeattention increases only by 1.5 and 3 percentage points.Thismeans thatamong thedifferent groupsof thepublic,the General Public has the lowest ability to lead theagenda of members of Congress. The effect of a per-manent increaseof attention by the general public (blackgeneral public estimates in the two left panels) is ofa similar magnitude to a one-time attention increase byparty supporters (Democratic supporter gray estimate inthe left panel and Republican supporter gray coefficientin the second panel from the left). Moreover, given thelowattention thegeneralpublicpays topolitics, attentionshifts of this magnitude (10 percentage points) are un-likely to take place. Nevertheless, when comparing theresults of fitting the same model only to data from 2013with data only from 2014, we interestingly find thatpoliticians are slightly more responsive to the generalpublic during election year (2014) than during a non-election year (although even during an election yearpoliticiansarealsomore likely to followshifts inattentionby party supporters and the attentive public).22

Overall, the results show thatpoliticians aremore likelyto follow changes in issue attention distribution by theirown party supporters than to attentive voters, and thatthey rarely follow the issuepriorities of the general public.

Responsiveness and Issue Relevance

If members of Congress have an interest in being re-sponsive to specific groups of constituents, then weexpect that (H6) they should be particularly interestedin responding to changes in attention involving issuesthat are salient to these groups. To test this hypothesis,we first need to estimate by how much each group ledthe attention that all other groups paid to each separatepolitical topic. To do so, we relax our assumption thatthe ability of one group to lead the agenda of the othersis constant across issues, and we model the data ina different way. In the previous model, we includedtopic-fixed effects (aj). In this section, we instead esti-mate 46 separate VAR models, one for each politicalissue. We include the same endogenous variables intothemodel, again apply a logit transformation to all timeseries and use the same seven-lag structure. Then, foreach of the VAR issue models, we calculate 15-daycumulative IRFs capturing how a one-time 10 per-centage point increase in attention by a specific grouppredicts the attention of the other groups.

Figure 4 shows the results (15-day IRFs) for each ofthese 46 VAR models. Each panel reports how thegroups in the panel titles are predicted to respond tochanges in attention by the other groups: the circlesrepresent the predicted effects (with lines representinga 95% confidence intervals), the colors of the circlesshow thegroup towhich theyare reacting, and the labelson the y-axis (the row labels) indicate the specific issue.To avoid overcrowding the plot, in the two left-mostpanels we only show the ability of the public groups tolead the expressed agenda of Democrats and Repub-licans inCongress, and in the four panels on the rightweshow the reverse effects, the ability of members ofCongress to lead public issue attention.Weonly includethe predicted effects for issues where the confidenceintervals do not cross zero.23

The issuesare sortedbasedon thepredicted impactofDemocratic Supporters on Democrats in Congress:from the issues with the largest estimated impact to theissues with the smallest. Among the top rows, we findissues such as healthcare reform (“Affordable CareAct,” row 1), gun violence (row 2), and minimum wage(row 7). A one-time 10-percentage-point increase onthese issues by Democratic supporters is predicted toincrease the cumulative attention that Democraticmembers of Congress pay to them by 2, 1.5, and 1percentage points.

In the second panel from the left, we see the issues inwhich Republican supporters more strongly led theagendaofRepublicans inCongress by lookingat the redcircles.Thediscussions around the IRS scandal (row35)that took place around mid-2013 and the imple-mentation of the Affordable Care Act and its websiteproblems (row 23)were the issues onwhichRepublicanmembers of congress seemed to follow their supportersthe most. In both cases, a one-time 10 percentage pointshift is predicted to translate into an increase of cu-mulative attention of about 2 percentage points 15 days

22 See Online Appendix E.1. 23 In Online Appendix C, we show predicted effects for all issues.

Pablo Barbera et al.

894

Dow

nloa

ded

from

htt

ps://

ww

w.c

ambr

idge

.org

/cor

e. IP

add

ress

: 54.

39.1

06.1

73, o

n 25

Aug

202

0 at

07:

56:1

1, s

ubje

ct to

the

Cam

brid

ge C

ore

term

s of

use

, ava

ilabl

e at

htt

ps://

ww

w.c

ambr

idge

.org

/cor

e/te

rms.

htt

ps://

doi.o

rg/1

0.10

17/S

0003

0554

1900

0352

later. The effects represented by gray circles in the left-most panel indicate that social security (row 4) is theissue on which the attentive public was able to lead thediscussion of Democratic members the most, and in thesecond panel, we see that they particularly led the at-tention thatRepublicanmembers paid to the protests inUkraine and Venezuela (row 30). The effects repre-sented by orange circles in the two left panels indicatethat discussions by Democratic and Republican mem-bers of congress on student debt (row 16) and the use ofmilitary force in Syria (row 17), respectively, wereamong the ones that the general public was able topositively lead the most.

To test the last of our hypothesis, that members ofCongress will bemore likely to follow shifts in attentionon issues towhich constituents paymore attention (H6),we build a measure of group issue relevance by calcu-lating the average daily attention each group paid toeach topic during the 113th Congress (these averagesare displayed in Figure 1). By taking the average, weintend to focus less on howmuch attention a group paidto a given issue at a particular point in time and tocapture instead how important the issue was for thatspecific group in general.

With this measure of average attention and theestimates from Figure 4 in hand, we can now move to

FIGURE 4. Predicted Issue Responsiveness Across Issues and Groups (15 Day IRFs)

Note: Thecoefficients (with 95%confidence intervals) indicate (in percentagepoints) howmuchmore/lesscumulativeattention thegroups inthepanel titlespaid to the issue in they-axisasa result of agroup (identifiedby thecolor) increasing thecumulativeattention to thesame issueby 10percentagepoints 15 days ago.Only coefficients not crossing zero havebeen included. The two left-most panels show the influence ofthe public onMembers of Congress. The four right-most panels show the influence of Democratic andRepublicanmembers of Congress onthe public. Versions of this figure that also show the coefficients crossing zero are available in Online Appendix C.

Who Leads? Who Follows?

895

Dow

nloa

ded

from

htt

ps://

ww

w.c

ambr

idge

.org

/cor

e. IP

add

ress

: 54.

39.1

06.1

73, o

n 25

Aug

202

0 at

07:

56:1

1, s

ubje

ct to

the

Cam

brid

ge C

ore

term

s of

use

, ava

ilabl

e at

htt

ps://

ww

w.c

ambr

idge

.org

/cor

e/te

rms.

htt

ps://

doi.o

rg/1

0.10

17/S

0003

0554

1900

0352

a direct test of H6 by examining correlations betweenthe two. Accordingly, in Figure 5 we plot on the x-axisthe average daily attention paid to each issue by eachgroup of the public (see panel titles). In the y-axis, weplot the cumulative attention members of Congress arepredicted to pay to each issue as result of the groups inthe panel title increasing their attention by 10 per-centage points 15 days ago (15 day cumulative IRFs).Each dot is a single predicted response and the linesaround them represent 95% confidence intervals. Thefour top panels show the predicted response of Dem-ocratic members of Congress, whereas the bottom onesillustrate the predicted reaction of Republicanlawmakers.

We find support for the issue relevance hypothesis(H6) only as it relates to the Supportermodel. In the topright panel, we observe that changes in attention byDemocratic supporters have a larger effect on theagenda of Democrats in Congress when they involveissues Democratic supporters deem relevant (such asgun violence). In the second from the right bottompanel, we also observe a similar pattern for Repub-licans, with Republican supporters being more likely tolead the expressed agenda of Republicans in Congresson issues that are important to them (such as the dis-cussion around the Internal Revenue Service—IRS).Thus, members of congress appear to behave as if theyare more likely to pay attention to the views of theirsupporters on issues their supporters care more aboutthan on issues they care less.

In Figure 5, we see no support for the issue relevancehypothesis as it relates to the Attentive and Downsianmodels. On average, Democrats in Congress are morelikely to follow changes in attention by the attentivepublic (including supporters of the other party) onissues that are relevant to these groups (two middlepanels on the top).However, these positive correlationsare of a very small magnitude. For Republicans inCongress, we do not even see a positive correlationbetween how much they follow shifts in attention onparticular topics and the average attention the attentivepublic and supporters of the other party paid to theissues (flat yellow and blue lines at the bottom). Finally,we observe no correlation between how much thegeneral public leads politicians’ agenda and the amountof average attention the general public devotes to anygiven issue. Overall, these results validate the strongfindings from Figure 2 in favor of the Supporter model,as well as the lack of evidence in support of theDownsian model of responsiveness.

The Role of the Media

Wenow turn to an evaluation of the role played bynewsmedia in mediating the dynamics identified in this pa-per.Our data andmethods allow us to examine in detailwhether the mass media is equally likely to lead, and beled by, politicians and the public—or whether, in con-trast, the media strengthens the voice of some groupsand increases their ability to lead political agendas.

FIGURE 5. Correlation Between Public Issue Relevance and the Ability of the Public to Set PoliticalAgendas

Note: The x-axis indicates theaverageattention thegroups in the toppanel titles paid to eachpolitical topic during the113thCongress.The y-axis indicateshowmuchmore/less cumulativeattentionDemocrats (top four panels) andRepublicans inCongress (bottom four panels) paidto these topicsasa result of thegroups in the toppanel titles increasing theattention to the topicby10percentagepoints15daysago.Eachdotrepresentsadifferent political issueand the linesaround thedots represent 95%confidence intervals.Rowsaresortedby the largest effect ofDemocrats in Congress (left panel).

Pablo Barbera et al.

896

Dow

nloa

ded

from

htt

ps://

ww

w.c

ambr

idge

.org

/cor

e. IP

add

ress

: 54.

39.1

06.1

73, o

n 25

Aug

202

0 at

07:

56:1

1, s

ubje

ct to

the

Cam

brid

ge C

ore

term

s of

use

, ava

ilabl

e at

htt

ps://

ww

w.c

ambr

idge

.org

/cor

e/te

rms.

htt

ps://

doi.o

rg/1

0.10

17/S

0003

0554

1900

0352

As a first cut, Table 4 confirms previous work in-dicating that media coverage reflects both politicians’and the public’s issue priorities. The table displayscorrelations between the distribution of attention toissues by the media and that of the other groups understudy. We find a particularly strong relationship be-tween media issue attention and the issue attention ofpolitically engaged Americans—that is, party sup-porters and the attentive public. Less substantial, butstill very strong, relationships exist between the issueattention of mass media and that of members of Con-gress and the general public.

These correlations, of course, do not provide in-formation about the directions of these relationships.For a clearer picture of the role of the media, we takea closer look at the IRF coefficient estimates regardingthe media originally displayed in the final column andbottom row of Figure 2, plotting them in more detail inFigure 6. In this figure’s left panel,media coverage is thedependent variable. Plotted here are the estimatedimpacts of changes in the attention to issues given bypoliticians and the public on the attention given to theseissues by media. Here, we observe that demand-sideforces (namely, the priorities of the most politicallyengagedAmericans) are strongerpredictors ofwhat themedia covers than supply-side forces (i.e., the prioritiesofmembers ofCongress). These results corroborate the

argument that media outlets are particularly likely tofollow shifts in attention by the public due to marketpressures (Anand and Peterson 2000; Webster andKsiazek 2012) and suggest that the effort expended bylawmakers to raise the salience of their favored issues isrelatively less influential.

In the right panel of Figure 6, media coverage is theindependent variable. This panel displays the estimatedimpacts of changes in media issue attention on the at-tention given to these issues by politicians and thepublic. Shifts in issueattentionbymediaoutletshave thestrongest impact on the issue agendas of members ofCongress andof party supporters.Notably, in each case,the power of shifts in media attention to predict sub-sequent shifts inattentionamongall audiences is greaterthan the reverse, confirming that media outlets playa crucial role in leadingpolitical attention (Soroka 2002;Walgrave, Stuart, and Nuytemans 2008; Walgrave andVan Aelst 2006). Finally, the media effects in Figure 6suggest that news media contribute to the promotion ofa Supporter responsivenessmodel. Not all groups of thepublic are equally likely to lead the issues covered bymedia outlets. The “voice” of Democratic and Re-publican supporters is stronger than the voice of theattentive citizens and the general public. This is par-ticularly relevant given that, as we observe in the rightpanel in Figure 6, the issues covered by the media arestrong predictors of the subsequently expressed issueagenda of members of Congress.

DISCUSSION AND CONCLUSIONS

It is well known inAmerican politics that politicians andthe public tend to pay attention to the same politicalissues (Jones and Baumgartner 2004), but due to datalimitations, the question of who leads whom has pre-viously beenunanswered (Burstein 2003). In this paper,we have contributed to answering this open question bycharacterizing the agenda of members of Congress andtheir constituents using latent topicmodeling applied tothe text of the tweets they sent between January 2013

TABLE 4. Correlation in Issue AttentionBetween Media Outlets and the Other Groups ofAnalysis

Group Media

Democrats in Congress 0.52Republicans in Congress 0.63Attentive public 0.74Democratic supporters 0.79Republican supporters 0.79General public 0.55

FIGURE 6. Predicted Media Effects

Note: The effects (with 95%) in the left panel indicate how much the media outlets increased their attention to a given issue (in percentagepoints) 15 days after the groups in the y-axis increased their attention to the same issue by 10 percentage points. The coefficients in the rightpanel indicates theviceversaeffects, howmuch thegroups in they-axis increase their attention toan issue15daysafter themedia increasedthe attention to the same issue by 10 percentage points.

Who Leads? Who Follows?

897

Dow

nloa

ded

from

htt

ps://

ww

w.c

ambr

idge

.org

/cor

e. IP

add

ress

: 54.

39.1

06.1

73, o

n 25

Aug

202

0 at

07:

56:1

1, s

ubje

ct to

the

Cam

brid

ge C

ore

term

s of

use

, ava

ilabl

e at

htt

ps://

ww

w.c

ambr

idge

.org

/cor

e/te

rms.

htt

ps://

doi.o

rg/1

0.10

17/S

0003

0554

1900

0352

and December 2014 (113th Congress). In doing so, wehave been able to create fine-grained political andpublic agendameasures and to studynot only the extentto which members of Congress follow shifts in issueattention by their constituents when deciding whatissues to discuss, but also to adjudicate between threecompeting models of political responsiveness: whetherpublic representatives follow changes in attention bytheir party supporters, the attentive public, or thegeneral public.

Wemodeled how public and political agendas predicteach other using a VAR model accounting for endog-enous and media effects. First, we found a political re-sponsiveness dynamic to be in place during the period ofanalysis. The public was not only able to lead theexpressed agenda of members of Congress, but themagnitude of this phenomenon was greater than thatassociatedwithpoliticians’ ability to leadpublic agendas:anattentionshift frompartysupportersprecededa largershift by members of Congress than the attention shiftfrom party supporters following an attention shift bymembers of Congress. Moreover, we found strongersupport for some responsiveness models than others.Our findings suggest that members of Congress aremainly responsive to changes in attention allocation byparty supporters and, to a lesser extent, attentive publics.The findings also suggest that mainstream media pro-mote similar dynamics: they are particularly likely tofollow the issuepreferencesofparty supporters, and theyare likely to lead the issue agenda of members of Con-gress. In addition, we observed Democrats andRepublicans in Congress to particularly follow partysupporterson issues thatare relevant to them.Finally,wefound very little empirical support for the claim thatpoliticians are responsive to the general public.

These issue responsiveness findings align with theliterature on policy and policy preference re-sponsiveness that shows that political and policyagendas in the United States are mainly driven by thepriorities of strong partisans (Clinton 2006; Kastellecet al. 2015; Shapiro et al. 1990). The study also supportsthe claim that, due to existing representation and re-sponsiveness dynamics, political agendas are morepolarized than is the American public (Grimmer 2013).While others show that this is in part a function ofgeographic sorting and an increasing number of clearlypartisan districts (Bishop and Cushing 2008; Grimmer2013), we show that low political attention by thegeneral public and a higher media coverage of partisanissue preferences is also in part to blame. This hasimportant normative implications for democratic poli-tics, as it could be an indirect factor contributing topolitical polarization.

Our analysis is limited to the 113th Congress, but weargue that our findings are likely to be generalizable tothe current context. Social media usage by members ofCongress was already almost universal in 2013. In ad-dition, data from the Pew Research Center show thatTwitter penetration among online US adults hasremained around 20% for the last five years (Smith andAnderson2018).The fact thatTwitter hasbecomemorecentral in US politics (Gainous and Wagner 2013),

particularly after Trump was elected President, meansthatwewould expect ourfindings tobemorevalid todayand lead tomoreprecise estimates.However, one resultthat may not hold today may be the notable partisanasymmetry we observe regarding the relationship be-tween legislators and the media, which appears to bestronger for Republicans than for Democrats. Thispattern could be a function of the political context, andin particular of which party is in the opposition, or itcould be related to structural factors, such as theasymmetric fragmentation of the media system-—currently larger on the left than on the right in theUnited States.

Overall, we illustrated how researchers can use socialmedia communications to uncover agenda setting andresponsiveness dynamics. Due to space constraints wehad to limit the scope of our analysis, but other basicquestions can be examined using this method. For ex-ample, is the President able to set political and publicagendas? Previous research shows that the President’sability is limited (Edwards and Wood 1999), but morerecent studies argue that this patternmay have changedin the last few years (Lawrence and Boydstun 2017;Wells et al. 2016)—a finding that may be worth revis-iting in the era of Trump’s presidency. Do politiciansrunning in safe versus marginal districts respond todifferent types of constituents? Do politicians responddifferently to constituents’ issue priorities depending onthe issues they own?Andhowwould these results differacross institutional or political contexts? In particular,we might have reason to expect higher levels of re-sponsiveness in countries with higher levels of politicalcontestation (Hobolt and Klemmensen 2008).

Another accessible topic of study is issue re-sponsiveness at the state level. Existing responsivenessresearch in the United States studies how the issuesdiscussedbyFederal political elites are shapedbypublicissue and policy preferences (Erikson, Mackuen, andStimson 2002; Jones and Baumgartner 2004; Page andShapiro 1983; Stimson,Mackuen, andErikson 1995). Arelevant number of political decisions, however, aremade at the state level. Are state policy makers re-sponsive to their constituents? What type of con-stituents? Do federal agendas influence politicaldiscussions at the state level? Do we see differentialresponsiveness dynamics across states? And if so, why?In addition, given a longer time period, one couldcombine the fine-grained temporal measure thatTwitter data offer with the curated topics of the PolicyAgenda Project to determine who leads and who fol-lows on each of the 19 issues the project has defined(Jones and Baumgartner 2004). Our hope is that boththe findings and methods introduced here can serve asa springboard for research into these and other im-portant topics related to political representation in thefuture.

SUPPLEMENTARY MATERIAL

To view supplementary material for this article, pleasevisit https://doi.org/10.1017/S0003055419000352.

Pablo Barbera et al.

898

Dow

nloa

ded

from

htt

ps://

ww

w.c

ambr

idge

.org

/cor

e. IP

add

ress

: 54.

39.1

06.1

73, o

n 25

Aug

202

0 at

07:

56:1

1, s

ubje

ct to

the

Cam

brid

ge C

ore

term

s of

use

, ava

ilabl

e at

htt

ps://

ww

w.c

ambr

idge

.org

/cor

e/te

rms.

htt

ps://

doi.o

rg/1

0.10

17/S

0003

0554

1900

0352

Replication materials can be found on Dataverse at:https://doi.org/10.7910/DVN/AA96D2.

REFERENCES

Adler, Scott, and John Wilkerson. 2012. Congress and the Politics ofProblem Solving. New York: Cambridge University Press.

Aldrich, John H. 1995.Why Parties? The Origin and Transformationof Political Parties in America. Chicago: The University of ChicagoPress.

Anand, N., and Richard A. Peterson. 2000. “When Market In-formation Constitutes Fields: Sensemaking of Markets in theCommercial Music Industry.”Organization Science 11 (3): 270–84.

Arnold, R. Douglas. 1990. The Logic of Congressional Action. NewHaven: Yale University Press.

Baumgartner, Frank R., and Bryan D. Jones. 1993. Agendas andInstability in American Politics. Chicago: University of ChicagoPress.

Bawn, Kathleen, Martin Cohen, David Karol, Seth Masket, HansNoel, and JohnZaller. 2012.“ATheoryof Political Parties:Groups,Policy Demands and Nominations in American Politics.” Per-spectives on Politics 10 (3): 571–97.

Birkland, Thomas A. 1998. “Focusing Events, Mobilization, andAgenda Setting.” Journal of Public Policy 18 (1): 53–74.

Bishop, Bill, and Robert G. Cushing. 2008. The Big Sort: Why theClustering ofLike-MindedAmerica IsTearingUsApart. NewYork:Houghton Mifflin.

Blei, David M., Andrew Y. Ng, and Michael I. Jordan. 2003. “LatentDirichletAllocation.”The JournalofMachineLearningResearch3:993–1022.

Boydstun, Amber E. 2013.Making the News: Politics, the Media, andAgenda Setting. Chicago: University of Chicago Press.

Burstein,Paul. 2003. “The ImpactofPublicOpiniononPublic Policy:AReview and an Agenda.” Political Research Quarterly 56 (1): 29–40.

Burstein, Paul. 2014.AmericanPublicOpinion,Advocacy, andPolicyin Congress: What the Public Wants and what it Gets. New York:Cambridge University Press.

Campbell, Angus, Philip E. Converse, Warren E. Miller, and DonaldE. Stokes. 1960. The American Voter. New York: John Wiley.

Campbell, James E., Bryan J. Dettrey, andHongxingYin. 2010. “TheTheory of ConditionalRetrospectiveVoting:Does the PresidentialRecord Matter Less in Open-Seat Elections?” The Journal ofPolitics 72 (4): 1083–95.

Canes-Wrone, Brandice, and Kenneth W. Shotts. 2004. “The Con-ditional Nature of Presidential Responsiveness to PublicOpinion.”American Journal of Political Science 48 (4): 690–706.

Casas, Andreu, and David Morar. 2015. “Different Channel, SameStrategy? Filling Empirical Gaps in congress Literature.” In Pro-ceedings of the Annual Meeting of the American Political ScienceAssociation, eds. Layna Mosley and Alvin Bernard Tillery. SanFrancisco: American Political ScienceAssociation. 1–21. Availableat SSRN: https://ssrn.com/abstract=3395307.

Caughey, Devin, and Christopher Warshaw. 2018. “Policy Prefer-ences and Policy Change: Dynamic Responsiveness in the Amer-ican States, 1936–2014.”American Political Science Review 112 (2):249–66.

Chang, Jonathan,SeanGerrish,ChongWang, JordanL.Boyd-graber,and David M. Blei. 2009. “Reading Tea Leaves: How HumansInterpret Topic Models.” In Advances in Neural InformationProcessing Systems. Vol. 22, eds.: Yoshua Bengio, Dale Schuur-mans, John Lafferty, Chris Williams, and Aaron Culotta. Van-couver, BC: Curran Associates, Inc., 288–96.

Chen, Kaiping, Nathan Lee, and William Marble. 2018. “Do ElectedOfficials Listen to Constituents on Social Media? Survey Evidencefrom Local Politicians in the United States.” Working paper.Available at SSRN: https://ssrn.com/abstract=3251651 or http://dx.doi.org/10.2139/ssrn.3251651.

Chi, Feng, and Nathan Yang. 2011. “Twitter Adoption in congress.”Review of Network Economics 10 (1): 1–46.

Clinton, Joshua D. 2006. “Representation in congress: Constituentsand Roll Calls in the 106th House.” The Journal of Politics 68 (2):397–409.

Converse, Philip E. 2006. “The Nature of Belief Systems in MassPublics.” Critical Review 18 (1–3): 1–74.

Delli Carpini, Michael X., and Scott Keeter. 1996. What AmericansKnow about Politics and Why it Matters. New Haven: Yale Uni-versity Press.

de Vreese, Claes H., and Hajo Boomgaarden. 2006. “News, PoliticalKnowledge and Participation: The Differential Effects of Newsmedia Exposure on Political Knowledge and Participation.” ActaPolitica 41 (4): 317–41.

Downs, Anthony. 1957. An Economic Theory of Democracy. NewYork: Harper and Row.

Edwards, George C., and B. Dan Wood. 1999. “Who InfluencesWhom? The President, congress, and the media.” American Po-litical Science Review 93 (2): 327–44.

Egan, Patrick J. 2013.PartisanPriorities:How IssueOwnershipDrivesand Distorts American Politics. New York: Cambridge UniversityPress.

Enders, Walter, and Todd Sandler. 1993. “The Effectiveness ofAntiterrorism Policies: A Vector-Autoregression-InterventionAnalysis.” American Political Science Review 87 (4): 829–44.

Erikson,RobertS.,MichaelB.Mackuen, and JamesA.Stimson. 2002.The Macro Polity. New York: Cambridge University Press.

Evans, Heather K., Victoria Cordova, and Savannah Sipole. 2014.“Twitter Style: An Analysis of How House Candidates UsedTwitter inTheir2012Campaigns.”PS:PoliticalScience&Politics47(2): 454–62.

Feezell, Jessica T. 2018. “Agenda Setting through Social media: TheImportance of IncidentalNewsExposure and Social Filtering in theDigital Era.” Political Research Quarterly 71 (2): 482–94.

Fenno, Richard F. 1978. Home Style: House Members in Their Dis-tricts. London: Longman.

Freeman, John R., John T. Williams, and Tse-min Lin. 1989. “VectorAutoregression and the Study of Politics.” American Journal ofPolitical Science 33 (4): 842–77.

Gainous, Jason, and KevinM.Wagner. 2013. Tweeting to Power: TheSocial Media Revolution in American Politics. New York: OxfordUniversity Press.

Geer, J.G. 1996. From Tea Leaves to Opinion Polls: A Theory ofDemocratic Leadership. New York: Columbia University Press.

Gerber, Alan S., Dean Karlan, and Daniel Bergan. 2009. “Does themedia Matter? A Field experiment Measuring the Effect ofNewspapersonVotingBehavior andPoliticalOpinions.”AmericanEconomic Journal: Applied Economics 1 (2): 35–52.

Gerber, Elisabeth R., and Rebecca B. Morton. 1998. “PrimaryElectionSystemsandRepresentation.” JournalofLaw,Economics,and Organization 14 (2): 304–24.

Gilens, Martin. 2012. Affluence and Influence: Economic Inequalityand Political Power in America. Princeton, New Jersey: PrincetonUniversity Press.

Gilens, Martin, and Benjamin I. Page. 2014. “Testing Theories ofAmerican Politics: Elites, Interest Groups, and Average Citizens.”Perspectives on Politics 12 (3): 564–81.

Gitlin, Todd. 1980. The Whole World Is Watching: Mass Media in theMaking & Unmaking of the New Left. Berkeley, CA: University ofCalifornia Press.

Golbeck, Jennifer, Justin M. Grimes, and Anthony Rogers. 2010.“TwitterUse by theUSCongress.” Journal of theAmerican Societyfor Information Science and Technology 61 (8): 1612–21.

Graber, Doris A. 1997. Mass Media and American Politics. Wash-ington, D.C.: CQ Press.

Grimmer, Justin. 2013. “Appropriators Not Position Takers. theDistorting Effects of Electoral Incentives on Congressional Rep-resentation.”American Journal of Political Science. 57 (3): 624–42.

Grimmer, Justin, and Brandon M. Stewart. 2013. “Text as Data: ThePromise and Pitfalls of Automatic Content Analysis Methods forPolitical Texts.” Political Analysis 21 (3): 267–97.

Habel, Philip D. 2012. “Following the Opinion Leaders? The Dy-namics of Influence Among media Opinion, the Public, and Poli-ticians.” Political Communication 29 (3): 257–77.

Hemphill, Libby, Jahna Otterbacher, and Matthew Shapiro. 2013.“What’s Congress Doing on Twitter?” In Proceedings of the 2013Conference on Computer Supported Cooperative Work, eds. AmyBruckman, Scott Counts, Cliff Lampe, and Loren Terveen. NewYork: ACM, 877–86.

Who Leads? Who Follows?

899

Dow

nloa

ded

from

htt

ps://

ww

w.c

ambr

idge

.org

/cor

e. IP

add

ress

: 54.

39.1

06.1

73, o

n 25

Aug

202

0 at

07:

56:1

1, s

ubje

ct to

the

Cam

brid

ge C

ore

term

s of

use

, ava

ilabl

e at

htt

ps://

ww

w.c

ambr

idge

.org

/cor

e/te

rms.

htt

ps://

doi.o

rg/1

0.10

17/S

0003

0554

1900

0352

Hemphill, Libby, and Andrew J. Roback. 2014. “Tweet Acts: HowConstituents Lobby congress via Twitter.” In Proceedings of the17th ACM Conference on Computer Supported Cooperative Work& Social Computing, eds. Susan Fussell, Wayne Lutters, MeredithRingel Morris, and Madhu Reddy. New York: ACM, 1200–10.

Hibbing, John R., and Elizabeth Theiss-Morse. 2002. Stealth De-mocracy:Americans’Beliefs aboutHowGovernment ShouldWork.Cambridge, UK: Cambridge University Press.

Hobolt, Sara, and Robert Klemmensen. 2008. “Government Re-sponsiveness and Political Competition in Comparative Perspec-tive.” Comparative Political Studies 41 (3): 309–37.

Holbrook, Thomas M., and Scott D. McClurg. 2005. “The Mobili-zation of Core Supporters: Campaigns, Turnout, and ElectoralComposition in United States Presidential Elections.” AmericanJournal of Political Science 49 (4): 689–703.

Hong, Liangjie, and Brian D. Davison. 2010. “Empirical Study ofTopic Modeling in Twitter.” In Proceedings of the First Workshopon Social Media Analytics, eds. PremMelville, Jure Leskovec, andFoster Provost. New York: ACM, 80–8.

Iyengar, Shanto, and Donald R. Kinder. 2010. News that Matters:Television and American Opinion. Chicago: University of ChicagoPress.

Jacobs, Lawrence R., and Robert Y. Shapiro. 1997. “TheMyth of thePandering Politician.” The Public Perspective 8: 3–5.

Jacobs, Lawrence R., and Robert Y. Shapiro. 2000. Politicians Don’tPander: Political Manipulation and the Loss of Democratic Re-sponsiveness: University of Chicago Press.

Jones, Bryan D. 2004. Reconceiving Decision-Making in DemocraticPolitics: Attention, Choice, and Public Policy. Chicago: The Uni-versity of Chicago Press.

Jones, Bryan D., and Frank R. Baumgartner. 2004. “Representationand Agenda Setting.” Policy Studies Journal 32 (1): 1–24.

Jones, Bryan D., and Frank R. Baumgartner. 2005. The Politics ofAttention: How Government Prioritizes Problems. Chicago: Uni-versity of Chicago Press.

Kastellec, Jonathan P., JeffreyR. Lax,MichaelMalecki, and JustinH.Phillips. 2015. “Polarizing the Electoral Connection: PartisanRepresentation in Supreme Court Confirmation Politics.” TheJournal of Politics 77 (3): 787–804.

Katz, Elihu, and Paul F. Lazarsfeld. 1955. Personal Influence: TheParty Played by People in the Flow of Mass Communications. NewYork: Free Press.

King, Gary, Benjamin Schneer, and Ariel White. 2017. “How theNews media Activate Public Expression and Influence NationalAgendas.” Science 358 (6364): 776–80.

Kingdon, John W. 2013. Agendas, Alternatives, and Public Policies.London, UK: Pearson Education Limited.

Ladd, Jonathan M., and Gabriel S. Lenz. 2009. “Exploiting a RareCommunication Shift to Document the Persuasive Power of theNewsmedia.”AmericanJournal ofPolitical Science53 (2): 394–410.

Lau, Richard R., and David P. Redlawsk. 2006. How Voters Decide:Information Processing in Election Campaigns. New York: Cam-bridge University Press.

Lawrence, Regina G., and Amber E. Boydstun. 2017. “What WeShould Really Be Asking about media Attention to Trump.” Po-litical Communication 34 (1): 150–3.

Lax, Jeffrey R., and Justin H. Phillips. 2011. “The Democratic Deficitin the States.”American Journal of Political Science 56 (1): 148–66.

Lupia, Arthur, Mathew D. McCubbins, and Lupia Arthur. 1998. TheDemocratic Dilemma: Can Citizens Learn what They Need toKnow? New York: Cambridge University Press.

McCombs, Maxwell E., and Donald L. Shaw. 1972. The Agenda-Setting Function of Mass media. Public Opinion Quarterly 36 (2):176–87.

Neundorf, Anja, and JamesAdams. 2018. “TheMicro-foundations ofParty Competition and IssueOwnership: TheReciprocal Effects ofCitizens’ Issue Salience and Party Attachments.”British Journal ofPolitical Science 48 (2): 385–406.

O’Connor, Brendan, Ramnath Balasubramanyan, Bryan R. Rout-ledge, and Noah A. Smith. 2010. “From Tweets to Polls: LinkingText Sentiment to Public Opinion Time Series.” In Fourth In-ternational AAAI Conference on Weblogs and Social Media, eds.Marti Hears, William Cohen, and Samuel Gosling. Menlo Park,CA: The AAAI Press, 122–8.

Page, Benjamin I. 1994. “Democratic Responsiveness? UntanglingtheLinks betweenPublicOpinion andPolicy.”PS: Political Scienceand Politics 27 (1): 25–9.

Page, Benjamin I., and Robert Y. Shapiro. 1983. “Effects of PublicOpinion on Policy.” American Political Science Review 77 (1):175–90.

Page, Benjamin I., and Robert Y. Shapiro. 1992.The Rational Public:Fifty Years of Trends in American’s Policy Preferences. Chicago:The University of Chicago Press.

Quinn, Kevin M., Burt L. Monroe, Michael Colaresi, Michael H.Crespin, and Dragomir R. Radev. 2010. “How to Analyze PoliticalAttention with Minimal Assumptions and Costs.” AmericanJournal of Political Science 54 (1): 209–28.

Russell, Annelise. 2018a. “The Politics of Prioritization: Senators’Attention in 140 Characters.” The Forum 16 (2): 331–56.

Russell,Annelise. 2018b. “USSenators onTwitter:Asymmetric PartyRhetoric in 140 Characters.” American Politics Research 46 (4):695–723.

Schattschneider,ElmerE. 1975.TheSemisovereignPeople:ARealist’sView of Democracy in America. New York: Dryden Press.

Schlozman, Kay Lehman, Sidney Verba, and Henry E. Brady. 2012.The Unheavenly Chorus: Unequal Political Voice and the BrokenPromise of American Democracy. Princeton, NJ: Princeton Uni-versity Press.

Shapiro, Catherine R., DavidW. Brady, Richard A. Brody, and JohnA. Ferejohn. 1990. “Linking Constituency Opinion and SenateVoting Scores: A Hybrid Explanation.” Legislative Studies Quar-terly 15 (4): 599–621.

Shapiro, MatthewA., and Libby Hemphill. 2017. “Politicians and thePolicyAgenda:DoesUseofTwitterby theUSCongressDirectNewYork Times Content?” Policy & Internet 9 (1): 109–32.

Shapiro, MatthewA., Libby Hemphill, and Jahna Otterbacher. 2012.“Doing what I Say: Connecting Congressional Social media Be-havior and Congressional Voting.” In Proceedings of the AnnualMeeting of the Midwest Political Science Association, eds. JonRogowski, Betsy Sinclair. Chicago: Midwest Political Science As-sociation, 1–32.

Shoemaker, Stephen, and Pamela Reese. 1996. Mediating the Mes-sage: Theories of Influences on Mass media Content. New York:Longman.

Sims, Christopher A. 1980. “Macroeconomics and Reality.” Econo-metrica 48 (1): 1–48.

Smith, Aaron, andMonicaAnderson. 2018. SocialMedia Use in 2018.Washington,DC:Pew Internet&AmericanLifeProject.RetrievedMarch 3, 2018.

Sniderman, Paul M., Richard A. Brody, and Phillip E. Tetlock. 1993.Reasoning and Choice: Explorations in Political Psychology. NewYork: Cambridge University Press.

Soroka, Stuart N. 2002. “Issue Attributes and Agenda-Setting bymedia, the Public, and Policymakers in Canada.” InternationalJournal of Public Opinion Research 14 (3): 264–85.

Soroka, Stuart N., and Christopher Wlezien. 2009. Degrees of De-mocracy: Politics, Public Opinion, and Policy. New York: Cam-bridge University Press.

Stimson, JamesA.,MichaelB.Mackuen, andRobert S.Erikson. 1995.“Dynamic Representation.” American Political Science Review 89(3): 543–65.

Sulkin, Tracy. 2005. Issue Politics in Congress. NewYork: CambridgeUniversity Press.

Tausanovitch, Chris, and Christopher Warshaw. 2014. “Represen-tation in Municipal Government.” American Political Science Re-view 108 (3): 605–41.

Walgrave, Stefaan, Stuart Soroka, and Michiel Nuytemans. 2008.“TheMassmedia’sPoliticalAgenda-SettingPower:ALongitudinalAnalysis of Media, Parliament, and Government in Belgium (1993to 2000).” Comparative Political Studies 41 (6): 814–36.

Walgrave, Stefaan, and Peter Van Aelst. 2006. “The Contingency ofthe Mass Media’s Political Agenda Setting Power: Toward a Pre-liminary Theory.” Journal of Communication 56 (1): 88–109.

Wallis,KennethF. 1987.“TimeSeriesAnalysisofBoundedEconomicVariables.” Journal of Time Series Analysis 8 (1): 115–23.

Webster, James G., and Thomas B. Ksiazek. 2012. “The Dynamics ofAudience Fragmentation: Public Attention in an Age of Digitalmedia.” Journal of Communication 62 (1): 39–56.

Pablo Barbera et al.

900

Dow

nloa

ded

from

htt

ps://

ww

w.c

ambr

idge

.org

/cor

e. IP

add

ress

: 54.

39.1

06.1

73, o

n 25

Aug

202

0 at

07:

56:1

1, s

ubje

ct to

the

Cam

brid

ge C

ore

term

s of

use

, ava

ilabl

e at

htt

ps://

ww

w.c

ambr

idge

.org

/cor

e/te

rms.

htt

ps://

doi.o

rg/1

0.10

17/S

0003

0554

1900

0352

Wells, Chris, Dhavan V. Shah, Jon C. Pevehouse, JungHwan Yang,Ayellet Pelled, Frederick Boehm, Josephine Lukito, ShreenitaGhosh,andJessicaL.Schmidt. 2016.“HowTrumpDroveCoverageto the Nomination: Hybrid Media Campaigning.” Political Com-munication 33 (4): 669–76.

Wilkerson, John, and Andreu Casas. 2017. “Large-scale Computer-ized Text Analysis in Political Science: Opportunities and Chal-lenges.” Annual Review of Political Science 20 (1): 529–44.

Wlezien, Christopher. 2004. “Patterns of Representation: Dynamicsof Public Preferences and Policy.” The Journal of Politics 66 (1):1–24.

Wood, B.Dan, and Jeffrey S. Peake. 1998. “TheDynamics of ForeignPolicy Agenda Setting.” American Political Science Review 92 (1):173–84.

Wouters, Ruud, and StefaanWalgrave. 2017. “Demonstrating Power:How Protest Persuades Political Representatives.” American So-ciological Review 82 (2): 361–83.

Wright, Gerald C. 1989. “Policy Voting in the US Senate: Who IsRepresented?” Legislative Studies Quarterly 14 (4): 465–86.

Zaller, John R. 1992. The Nature and Origins of Mass Opinion. NewYork: Cambridge University Press.

Who Leads? Who Follows?

901

Dow

nloa

ded

from

htt

ps://

ww

w.c

ambr

idge

.org

/cor

e. IP

add

ress

: 54.

39.1

06.1

73, o

n 25

Aug

202

0 at

07:

56:1

1, s

ubje

ct to

the

Cam

brid

ge C

ore

term

s of

use

, ava

ilabl

e at

htt

ps://

ww

w.c

ambr

idge

.org

/cor

e/te

rms.

htt

ps://

doi.o

rg/1

0.10

17/S

0003

0554

1900

0352